cd /news/large-language-models/json-mode-silently-stopped-all-7-ope… · home topics large-language-models article
[ARTICLE · art-83137] src=pub.towardsai.net ↗ pub= topic=large-language-models verified=true sentiment=↓ negative

JSON Mode Silently Stopped All 7 Open Models From Calling a Single Tool

A June 2026 paper from Focus Technology's AI lab found that JSON mode caused all seven open-weight models tested to stop calling tools entirely, with tool invocation dropping from 100% to 0% when a JSON schema was attached. The issue stems from a token bitmask that sets logits to -inf for non-schema tokens, and on a five-field extraction schema, the grammar writes 72.4% of response bytes with no free choice until byte 17.

read1 min views1 publishedAug 1, 2026
JSON Mode Silently Stopped All 7 Open Models From Calling a Single Tool
Image: Pub (auto-discovered)

Member-only story

Seven open-weight models. Seven suppression rates of exactly 100%. Tool invocation went from 100% to 0% the moment a JSON schema was attached to the same request, and not one line in any log said so.

That is the finding in a June 2026 paper out of Focus Technology’s AI lab, and the first time I read it I assumed it was a prompt bug. It is not. It is arithmetic, and once you see the arithmetic you cannot unsee it: response_format={"type":"json_schema"}

compiles your schema into a token bitmask, and the engine executes exactly one line that ends the argument.

logits[mask == 0] = -inf

Your model wanted to call a tool. The mask deleted the option before sampling ever ran.

I spent this morning rebuilding the mask math from scratch to check whether the claim survives contact with a calculator. It does, and the numbers are worse than the paper’s headline. On a routine five-field extraction schema, the grammar writes 72.4% of the response bytes and the model does not get a single free choice until byte 17. We stopped calling this “generation” a while ago and nobody updated the mental model.

── more in #large-language-models 4 stories · sorted by recency
── more on @focus technology 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/json-mode-silently-s…] indexed:0 read:1min 2026-08-01 ·